TECHNICAL REFERENCE
Deploy from AI agents (MCP)
Use the hosted DeployVera MCP server from Claude Code, Cursor, VS Code or Codex: install, sign-in, permissions, approvals and limits.
What it is
DeployVera runs a hosted MCP (Model Context Protocol) server at https://mcp.deployvera.com/mcp (Streamable HTTP). An AI agent in Claude Code, Cursor, VS Code or Codex that is connected to it can list the projects of the signed-in account, start a deployment, wait for the build, read build and runtime logs, explain a failed build, roll back and manage environment variables.
The agent acts with the permissions of the account or token that signed in, never more. Buying or changing plans, payments, billing, account settings and administration are not available to the agent and stay in the customer panel. The server is listed as com.deployvera/mcp in the MCP registry, as deployvera-mcp on npm, in the Claude connector directory (https://claude.ai/directory/connectors/deployvera) and on GitHub (https://github.com/deployvera/deployvera-mcp). The agent can create a project in a plan you already have, if it has a free application slot, and deploy it; creating a project always needs your approval, and ordering a plan stays in the panel.
Install: Claude Code
Run this once in a terminal. The first tool call opens a browser window for sign-in (OAuth).
claude mcp add --transport http deployvera https://mcp.deployvera.com/mcpInstall: Cursor
One-click install: https://cursor.com/install-mcp?name=deployvera&config=eyJ1cmwiOiJodHRwczovL21jcC5kZXBsb3l2ZXJhLmNvbS9tY3AifQ%3D%3D
Or add the server by hand to ~/.cursor/mcp.json:
{
"mcpServers": {
"deployvera": {
"url": "https://mcp.deployvera.com/mcp"
}
}
}Install: VS Code
Add the server to the MCP configuration of VS Code (mcp.json):
{
"servers": {
"deployvera": {
"type": "http",
"url": "https://mcp.deployvera.com/mcp"
}
}
}Install: Codex
Add the server to ~/.codex/config.toml and export the personal token (see the next section) as the environment variable named below:
[mcp_servers.deployvera]
url = "https://mcp.deployvera.com/mcp"
bearer_token_env_var = "DEPLOYVERA_TOKEN"Install: stdio client
Clients that can only start local processes can use the npm package instead of the hosted URL. The token variable it expects is described in the repository README.
npx -y deployvera-mcpSign-in: browser OAuth or personal token
Browser OAuth is the default: the client opens a DeployVera sign-in and consent screen, and the account owner chooses which permissions to grant. Connected apps can be disconnected in the panel under Settings > AI agent access.
A personal token is the alternative for clients without a browser flow. It is created in the panel under Settings > AI agent access, starts with dvp_, is shown only once, has a name, a chosen set of permissions and a validity period (30, 90 or 365 days), and can be revoked at any time; revoked access ends immediately. An account can hold at most 10 active tokens. Send it as Authorization: Bearer dvp_.... Treat it like a password and never commit it to a repository.
Permissions (scopes)
Every token and every OAuth login carries a subset of three scopes.
read: always on. Lists projects, databases, builds, logs, the names of environment variables, your own services, connectable repositories and domain status.deploy: start a deployment, cancel a build, roll back, create a project and manage a custom domain.env: write and delete environment variables. The agent can set a value but can never read a value back.
Approvals in the browser
Five actions always need approval from the account owner: rollback, delete_env_var, create_project, set_domain and remove_domain. The tool returns an approval link instead of acting. The owner opens it (deployvera.com/approve), signs in and confirms with the account password and two-factor code. Then the agent calls the same tool again with the approval id. An approval expires after 10 minutes and works for one action only; the agent cannot approve on its own, and check_approval only reports the state.
A token or login can also be set to require approval for every deploy. This is a per-token or per-login option in Settings > AI agent access.
Tools
The server exposes 21 tools.
list_projects,get_project: projects and their current state.list_databases,get_database_logs: managed databases (beta) and their logs.deploy: start a deployment of a connected project. Needs thedeployscope.get_build: build status;wait_seconds(0 to 120) makes the call wait for the build to finish.get_build_logs: build or runtime logs.explain_build_failure: a short explanation of why a build failed.cancel_build: stop a running build.rollback: return to an earlier release. Always needs approval.list_env_vars: names only, never values.set_env_vars: write variables. Needs theenvscope.delete_env_var: remove a variable. Always needs approval.check_approval: state of a pending approval.list_services: your own services with plan, status and free application slots. Shared services are not included.list_repositories: repositories you can deploy from in the connected GitHub or GitLab accounts (and Gitea where enabled). No tokens are returned.create_project: create a project in one of your own services from a Git repository. Always needs approval, and works only where the service has a free application slot; it never orders or changes a plan.start_domain_verification: returns the DNS records you must create for a custom domain. It changes no routing and needs no approval.set_domain: attach the custom domain once DNS is verified. Always needs approval.remove_domain: detach the custom domain. Always needs approval.check_domain: whether a domain is attached and whether each name currently points to the project.
Security limits
The agent cannot buy or change plans, make payments, change billing or account settings, or use any administrator function. Environment variable values are never returned to the agent. Requests are limited to 120 per minute. Every action runs under the signed-in account and is checked against the ownership of the project, exactly like the panel.
Even so, a connected agent can start deployments within the granted scopes. Grant only the scopes needed, prefer short validity periods, turn on approval for every deploy for production projects, and revoke tokens that are no longer used.
Typical workflow
After the code is pushed, ask the agent to deploy the project. It calls deploy, then get_build with wait_seconds set to 120. If the build fails it calls explain_build_failure and get_build_logs. A rollback is proposed only when asked, and waits for approval in the browser. Common build errors and fixes are in /docs/en/build-errors. A short walkthrough with example prompts is at https://deployvera.com/deploy-from-claude-code.